Skip to content

MoreLikeThis (MLT)

MoreLikeThis (MLT) surfaces documents that share significant terms with a seed document.

Solr Documentation: MoreLikeThis

Configuration Approaches

from taiyo.parsers import StandardParser
from taiyo.params import MoreLikeThisParamsConfig

parser = StandardParser(
    query="id:article-123",
    configs=[
        MoreLikeThisParamsConfig(
            fields=["title", "content"],
            min_term_freq=1,
            min_doc_freq=1,
            max_query_terms=20,
            boost=True,
            match_include=False,
        )
    ],
)

from taiyo.parsers import StandardParser

parser = StandardParser(query="id:article-123", rows=1).more_like_this(
    fields=["title", "content"],
    min_term_freq=1,
    min_doc_freq=1,
    max_query_terms=20,
    boost=True,
    match_include=False,
)

Basic Usage

parser = StandardParser(query=f"id:{target_id}", rows=1).more_like_this(
    fields=["title", "content"],
    min_term_freq=1,
    min_doc_freq=1,
    max_query_terms=20,
    boost=True,
    match_include=False,
)

response = client.search(parser, document_model=Article)

Key Parameters

MoreLikeThisParamsConfig(
    fields=["title", "content"],  # Fields to analyze for similarity
    min_term_freq=1,  # Minimum term frequency in the seed doc
    min_doc_freq=1,  # Minimum document frequency across the index
    max_doc_freq_pct=80,  # (Optional) Percentage threshold for common terms
    max_query_terms=20,  # Cap on interesting terms used for the query
    min_word_len=3,  # Ignore short tokens
    max_num_tokens_parsed=5000,  # Limit analysis for non term-vector fields
    boost=True,  # Boost by term relevance scores
    query_fields="title^2.0 content",  # Optional boosted query fields
    interesting_terms="details",  # Include term provenance in the response
    match_include=False,  # Exclude the seed document from results
)

Refer to the Apache Solr documentation for the full list of parameters and defaults.

Handling Results

MLT responses are mapped onto typed models in Taiyo to make downstream handling ergonomic:

  • SolrResponse.more_like_this: a mapping of source document IDs to SolrMoreLikeThisResult instances.
  • SolrMoreLikeThisResult.docs: parsed SolrDocument instances representing similar documents.
  • SolrMoreLikeThisResult.interesting_terms: optional per-document metadata when interesting_terms is set to list or details.

Example:

from taiyo.types import SolrMoreLikeThisResult

mlt_map: dict[str, SolrMoreLikeThisResult[Article]] = response.more_like_this or {}
related = mlt_map.get(target_id)

if related:
    print(f"Found {related.num_found} related docs")
    for doc in related.docs:
        print(doc.id, doc.title)

    if related.interesting_terms:
        print("Interesting terms:")
        if isinstance(related.interesting_terms, dict):
            for term, boost in related.interesting_terms.items():
                print(f"  {term}: {boost}")
        else:
            for term in related.interesting_terms:
                print(f"  {term}")

When working with interesting_terms="details", Solr returns a dictionary keyed by term with boost values. Taiyo surfaces those details in the SolrMoreLikeThisResult so you can inspect which tokens contributed to the similarity score.

Next Steps